Fast Classification Approach of Support Vector Machine with Privacy Preservation
Shitong Wang · Dianzi xuebao · 2012
The decision functions of various kernelized classification methods can be expressed as a combination of Support Vectors(SVs),i.e.SVM,which contain the individual privacy information,so this information will be released during detecting unknown samples.Meanwhile,the amount of SVs limits classification speed,i.e.the computational time complexity of SVM is O(|SVs|).For overcoming the above drawbacks,a fast classification approach of SVM with privacy preservation is proposed,which is based on the agent preimage of the center of minimum enclosing ball(MEB),and two preimage-finding methods are presented in this paper,called QP-based solution and direct solution respectively.Experimental results on UCI and PIE face image demonstrate that two drawbacks as above can not only be solved,but also the obtained effectiveness of the proposed method is competitive.